Can AI write good marketing copy? Part 4: The state of GenAI in 2025

Much has changed in the AI space since late 2023, when we published our three-part white paper on the capabilities of generative AI for content marketing. OpenAI’s new flagship model, GPT-4o, is clearly more capable than the one we used to generate the second part of our white paper. And now there are new generative AI models vying for users: Google’s Gemini, Microsoft’s Copilot, Anthropic’s Claude, xAI’s Grok and the Chinese-hedge-fund-backed DeepSeek, to name a few.

But after more than a year of model training and billions in expenditures, is AI any better of a solution for your marketing copy?

In some ways, yes. But the core issues that come with entrusting your content marketing to generative AI remain.

Refined and ready to serve

We’ve restricted this analysis to ChatGPT, largely because it’s the chatbot we used to produce that first white paper but also because it’s still the most popular and widely known. Just keep in mind that the downsides we’ll discuss here are inherent to the large language model (LLM) approach, which means they’re a factor whether you use ChatGPT, Gemini, Copilot, etc.

Compared to its predecessor, GPT-4o offers flashy new features and a refined user experience. It can work across text, images and audio. It can access the web reliably and even generate detailed reports with its deep research feature. New “reasoning” abilities (essentially, the model checking its own work) enable it to solve more complex problems. Plus, it can pack much more text into a single response; there’s no longer the need to prompt it five separate times to produce a five-page white paper, for instance.

But what about the quality of the text it generates? GPT-4o’s default “personality” is less formal and more expressive than previous models, so the text overall sounds more human. ChatGPT also seems better able to follow instructions established within a prompt or its customization options.

The perils of AI-produced marketing copy

Despite these enhancements, the opinion we gave in the final part of our white paper, where we compared the ChatGPT-generated white paper to our own, is unchanged. While ChatGPT and other chatbots show potential as tools for brainstorming, outlining and other pre-writing tasks, they still aren’t the replacement for professional writers that some might think.

When you process language and ideas at the unprecedented scale of today’s LLMs, you get a remarkably good imitation of human intelligence. But it’s just an imitation. There’s zero intelligence at work. Dubbing the latest generation of LLMs as “reasoning” models or inferring that the models are “thinking” during their processing is pure marketing hype. Even referring to chatbots like ChatGPT as “artificial intelligence” is misleading. In truth, generative AI models don’t do any “thinking” at all. Instead, they draw on their enormous datasets to determine the most likely response to any given prompt. It’s why observers such as Noam Chomsky have called ChatGPT and the like a “kind of super-autocomplete.”

What does that mean for your marketing content? When you ask ChatGPT to write you a blog, white paper or any other type of marcom collateral, what you get is a remix of text from the model’s training data. Ultimately, generative AI is incapable of producing anything new — making it a non-starter for thought leadership. And if you use AI for your content, you’ll sound the same as every other company that's also using AI, making it difficult to stand out. Your company will lack a unique and consistent brand voice, which is needed to help build trust and familiarity with your audience.

Hallucinations and AI fatigue

There’s also the fact that generative AI outputs can’t be trusted. AI “hallucinations” — that is, when generative AI confidently provides an incorrect answer — are a well-documented issue inherent to the LLM approach. AI-generated answers at the very top of Google search results have advised users to add glue to their pizza sauce and said running with scissors is good exercise. A study by the BBC further demonstrated the tendency of generative AI to get the facts wrong. When fed news stories and then asked questions about those stories, AI assistants produced answers that contained “significant issues of some form” 51 percent of the time. And the problem is only getting worse: in a benchmark test that involved asking questions about public figures, OpenAI’s latest available model, o4-mini, hallucinated 48 percent of the time compared to the previous model’s 33 percent hallucination rate.

The trust of your audience is critical to your business, which is why generative AI must be used with extreme care. Any error that slips by, whether caused by an AI hallucination or otherwise, will be attributed to your brand — and erode the trust you’ve worked hard to establish.

Trust is an issue with AI in another sense: what happens to the company information you share with these LLMs? ChatGPT and Gemini can now accept Word documents, Excel spreadsheets, PowerPoint presentations, PDFs and other files to inform their responses. But even where there’s the option of disabling the use of these sources for model training, it’s wise to be cautious if sharing proprietary or otherwise sensitive information.

Then there’s the matter of overexposure. AI is everywhere: in productivity suites like Office 365, operating systems like Windows 11 and iOS, and even Windows Notepad. And the signs of distrust are starting to show. “AI slop” is a label that’s increasingly being applied to AI-generated content. People have compiled lists of AI “tells”, drawing attention to words that AI tends to favour, like “delve,” “leverage” and “meticulously”. Even the em dash is being looked on as a sign of AI-written content. It all points to building skepticism toward AI-generated material. And when skeptical users suspect your content came from an AI, they’ll move on — and probably not with a higher opinion of your business.

Stand out with professional writing and editing services

Will generative AI ever be the ideal solution for your marketing writing needs? It’s impossible to say, but we feel it’s unlikely due to how LLMs like ChatGPT work. As it stands now: use with caution, if at all.

If you are using generative AI for your marketing copy, we can help ensure the final product is an asset, not a liability. We can do line-by-line copyediting for mechanics and fact checking, a stylistic update to ensure your brand is well represented, or a complete overhaul if the text isn’t what you had in mind. If your brand lacks a unique voice, we can help with that, too.

And if you’re not using generative AI? Of course, we can help there as well. Our expert team of professional writers and editors can produce the full range of marketing communications content: blogs, articles, web pages, brochures, white papers and more.

Let’s discuss your next project — whether it’s AI-generated or human-authored — and the value Ascribe can bring.